[PATCH] D35998: [TableGen] AsmMatcher: fix OpIdx computation when HasOptionalOperands is true

[Nirav Dave via Phabricator via llvm-commits]

niravd added a comment.

I suspect that the issue is variation in our cmake configurations. I just rebuilt from a clean build and got the same issues.

Try it with CMAKE_BUILD_TYPE=Debug and LLVM_ENABLE_ASSERETIONS=ON.

In https://reviews.llvm.org/D35998#828678, @alexandru.guduleasa wrote:

> Thank you for your reply.
>
> We applied the patch after the commit that you mentioned:
>
>   git log -n2 --oneline
>   db8a65e AsmMatcher: fix OpIdx computation when HasOptionalOperands is true
>   343f60c [LV] Avoid redundant operations manipulating masks 
>
>
> We confirmed that the generated code contains our modification:
>
>   cat AMDGPUGenAsmMatcher.inc | grep -e "DefaultsOffset"
>     unsigned DefaultsOffset[12];
>       DefaultsOffset[i] = NumDefaults;
>       OpIdx = *(p + 1) - DefaultsOffset[*(p + 1) - 1]; 
>
>
> After this, we explicitly ran the mentioned tests:
>
>   ./bin/llvm-lit ../llvm/test/MC/AMDGPU/gfx7_asm_all.s ../llvm/test/MC/AMDGPU/gfx8_asm_all.s ../llvm/test/MC/AMDGPU/gfx9_asm_all.s ../llvm/test/MC/AMDGPU/sopk.s ../llvm/test/MC/AMDGPU/vintrp.s ../llvm/test/MC/AMDGPU/vop1-gfx9.s ../llvm/test/MC/AMDGPU/vop2.s ../llvm/test/MC/AMDGPU/vop3-convert.s
>   -- Testing: 8 tests, 4 threads --
>   PASS: LLVM :: MC/AMDGPU/sopk.s (1 of 8)
>   PASS: LLVM :: MC/AMDGPU/vintrp.s (2 of 8)
>   PASS: LLVM :: MC/AMDGPU/vop1-gfx9.s (3 of 8)
>   PASS: LLVM :: MC/AMDGPU/vop2.s (4 of 8)
>   PASS: LLVM :: MC/AMDGPU/vop3-convert.s (5 of 8)
>   PASS: LLVM :: MC/AMDGPU/gfx7_asm_all.s (6 of 8)
>   PASS: LLVM :: MC/AMDGPU/gfx9_asm_all.s (7 of 8)
>   PASS: LLVM :: MC/AMDGPU/gfx8_asm_all.s (8 of 8)
>   Testing Time: 4.76s
>     Expected Passes    : 8
>
>
> Any suggestions?





https://reviews.llvm.org/D35998